M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

/** * Mimir Neo4j Graph Performance Benchmark Suite * * Benchmarks Neo4j graph database operations: * 1. Vector search queries (cosine similarity, hybrid search) * 2. Relationship traversal (1-3 hops) * 3. Batch operations (bulk writes) * 4. Complex queries (aggregations, subgraphs) * * Note: Uses pre-populated mock embeddings - does NOT test embedding generation * * Run with: npm run bench * Results are written to: testing/benchmarks/results/ */ import { describe, bench, beforeAll, afterAll } from 'vitest'; import neo4j, { Driver, Session } from 'neo4j-driver'; import fs from 'fs/promises'; import path from 'path'; import { fileURLToPath } from 'url'; const __filename = fileURLToPath(import.meta.url); const __dirname = path.dirname(__filename); // Configuration const NEO4J_URI = process.env.NEO4J_URI || 'neo4j://localhost:7687'; const NEO4J_USER = process.env.NEO4J_USER || 'neo4j'; const NEO4J_PASSWORD = process.env.NEO4J_PASSWORD || 'password'; const RESULTS_DIR = path.join(__dirname, 'results'); let driver: Driver; let session: Session; // Setup beforeAll(async () => { driver = neo4j.driver(NEO4J_URI, neo4j.auth.basic(NEO4J_USER, NEO4J_PASSWORD)); session = driver.session(); // Ensure results directory exists await fs.mkdir(RESULTS_DIR, { recursive: true }); // Create test data await setupTestData(); }); afterAll(async () => { await cleanupTestData(); await session.close(); await driver.close(); }); // Test data setup async function setupTestData() { console.log('Setting up test data...'); // Create 1000 nodes with mock embeddings await session.run(` UNWIND range(1, 1000) AS i CREATE (n:BenchmarkNode { id: 'bench-' + toString(i), type: 'test', content: 'Test content ' + toString(i), value: i, created: timestamp() }) `); // Add mock embeddings (1024 dimensions) to nodes // Using a deterministic pattern for reproducibility const batchSize = 100; for (let batch = 0; batch < 10; batch++) { const startId = batch * batchSize + 1; const endId = (batch + 1) * batchSize; await session.run(` MATCH (n:BenchmarkNode) WHERE toInteger(substring(n.id, 6)) >= $startId AND toInteger(substring(n.id, 6)) <= $endId SET n.embedding = [x IN range(0, 1023) | toFloat(toInteger(substring(n.id, 6)) + x) / 2048.0] `, { startId, endId }); } // Create relationships between nodes await session.run(` MATCH (n:BenchmarkNode) WHERE toInteger(substring(n.id, 6)) % 10 = 0 WITH n LIMIT 100 MATCH (m:BenchmarkNode) WHERE m.id <> n.id AND toInteger(substring(m.id, 6)) > toInteger(substring(n.id, 6)) WITH n, m LIMIT 500 CREATE (n)-[:RELATED_TO {weight: rand(), created: timestamp()}]->(m) `); console.log('Test data setup complete'); } async function cleanupTestData() { await session.run(` MATCH (n:BenchmarkNode) DETACH DELETE n `); } // ============================================================================ // 1. VECTOR SEARCH BENCHMARKS // ============================================================================ describe('Vector Search Performance', () => { bench('Vector similarity search (top 10)', async () => { // Generate a query embedding (mock) const queryEmbedding = Array.from({ length: 1024 }, (_, i) => (Math.sin(i) + 1) / 2); await session.run(` MATCH (n:BenchmarkNode) WHERE n.embedding IS NOT NULL WITH n, reduce(dot = 0.0, i IN range(0, size(n.embedding)-1) | dot + n.embedding[i] * $embedding[i] ) AS similarity ORDER BY similarity DESC LIMIT 10 RETURN n.id, similarity `, { embedding: queryEmbedding }); }, { iterations: 100 }); bench('Vector similarity search (top 25)', async () => { const queryEmbedding = Array.from({ length: 1024 }, (_, i) => (Math.cos(i) + 1) / 2); await session.run(` MATCH (n:BenchmarkNode) WHERE n.embedding IS NOT NULL WITH n, reduce(dot = 0.0, i IN range(0, size(n.embedding)-1) | dot + n.embedding[i] * $embedding[i] ) AS similarity ORDER BY similarity DESC LIMIT 25 RETURN n.id, similarity `, { embedding: queryEmbedding }); }, { iterations: 100 }); bench('Vector similarity search (top 50)', async () => { const queryEmbedding = Array.from({ length: 1024 }, (_, i) => Math.random()); await session.run(` MATCH (n:BenchmarkNode) WHERE n.embedding IS NOT NULL WITH n, reduce(dot = 0.0, i IN range(0, size(n.embedding)-1) | dot + n.embedding[i] * $embedding[i] ) AS similarity ORDER BY similarity DESC LIMIT 50 RETURN n.id, similarity `, { embedding: queryEmbedding }); }, { iterations: 100 }); bench('Vector search with filter', async () => { const queryEmbedding = Array.from({ length: 1024 }, (_, i) => (i % 2) / 2); await session.run(` MATCH (n:BenchmarkNode) WHERE n.embedding IS NOT NULL AND n.type = 'test' AND n.value > 100 WITH n, reduce(dot = 0.0, i IN range(0, size(n.embedding)-1) | dot + n.embedding[i] * $embedding[i] ) AS similarity ORDER BY similarity DESC LIMIT 25 RETURN n.id, similarity `, { embedding: queryEmbedding }); }, { iterations: 100 }); }); // ============================================================================ // 2. GRAPH TRAVERSAL BENCHMARKS // ============================================================================ describe('Graph Traversal Performance', () => { bench('Node lookup by ID', async () => { await session.run(` MATCH (n:BenchmarkNode {id: 'bench-500'}) RETURN n `); }, { iterations: 1000 }); bench('Node lookup by property index', async () => { await session.run(` MATCH (n:BenchmarkNode) WHERE n.type = 'test' AND n.value > 500 RETURN n LIMIT 100 `); }, { iterations: 500 }); bench('Relationship traversal (depth 1)', async () => { await session.run(` MATCH (n:BenchmarkNode)-[r:RELATED_TO]->(m) RETURN n.id, m.id, r.weight LIMIT 100 `); }, { iterations: 200 }); bench('Relationship traversal (depth 2)', async () => { await session.run(` MATCH path = (n:BenchmarkNode)-[:RELATED_TO*1..2]->(m) RETURN path LIMIT 50 `); }, { iterations: 100 }); bench('Relationship traversal (depth 3)', async () => { await session.run(` MATCH path = (n:BenchmarkNode)-[:RELATED_TO*1..3]->(m) RETURN path LIMIT 25 `); }, { iterations: 50 }); bench('Bidirectional traversal', async () => { await session.run(` MATCH (n:BenchmarkNode {id: 'bench-100'}) MATCH (n)-[:RELATED_TO]-(connected) RETURN connected LIMIT 50 `); }, { iterations: 200 }); bench('Find shortest path', async () => { await session.run(` MATCH (start:BenchmarkNode {id: 'bench-10'}) MATCH (end:BenchmarkNode {id: 'bench-990'}) MATCH path = shortestPath((start)-[:RELATED_TO*..5]-(end)) RETURN path `); }, { iterations: 50 }); }); // ============================================================================ // 3. BATCH OPERATIONS BENCHMARKS // ============================================================================ describe('Batch Operations Performance', () => { bench('Batch node creation (100 nodes)', async () => { await session.run(` UNWIND range(1, 100) AS i CREATE (n:TempBenchNode { id: randomUUID(), value: i, created: timestamp() }) `); // Cleanup await session.run(` MATCH (n:TempBenchNode) DELETE n `); }, { iterations: 50 }); bench('Batch node creation (1000 nodes)', async () => { await session.run(` UNWIND range(1, 1000) AS i CREATE (n:TempBenchNode { id: randomUUID(), value: i, created: timestamp() }) `); // Cleanup await session.run(` MATCH (n:TempBenchNode) DELETE n `); }, { iterations: 10 }); bench('Batch relationship creation (100 edges)', async () => { await session.run(` MATCH (n:BenchmarkNode) WITH collect(n) AS nodes UNWIND range(0, 99) AS i WITH nodes[i % size(nodes)] AS n1, nodes[(i + 1) % size(nodes)] AS n2 CREATE (n1)-[:TEMP_EDGE {weight: rand(), created: timestamp()}]->(n2) `); // Cleanup await session.run(` MATCH ()-[r:TEMP_EDGE]->() DELETE r `); }, { iterations: 50 }); bench('Batch property update (100 nodes)', async () => { await session.run(` MATCH (n:BenchmarkNode) WHERE n.value <= 100 SET n.updated = timestamp(), n.batch_updated = true `); // Cleanup await session.run(` MATCH (n:BenchmarkNode) WHERE n.batch_updated = true REMOVE n.updated, n.batch_updated `); }, { iterations: 100 }); bench('Batch delete with relationships (50 nodes)', async () => { // Create temp nodes with relationships await session.run(` UNWIND range(1, 50) AS i CREATE (n:TempDeleteNode {id: 'temp-' + i}) WITH n MATCH (m:BenchmarkNode) WHERE m.value = toInteger(n.id[5..]) CREATE (n)-[:TEMP_REL]->(m) `); // Delete them await session.run(` MATCH (n:TempDeleteNode) DETACH DELETE n `); }, { iterations: 50 }); }); // ============================================================================ // 4. COMPLEX QUERY BENCHMARKS // ============================================================================ describe('Complex Query Performance', () => { bench('Aggregation query', async () => { await session.run(` MATCH (n:BenchmarkNode)-[r:RELATED_TO]->(m) WITH n, count(r) AS outDegree, avg(r.weight) AS avgWeight WHERE outDegree > 0 RETURN n.id, outDegree, avgWeight ORDER BY outDegree DESC LIMIT 50 `); }, { iterations: 100 }); bench('Subgraph extraction', async () => { await session.run(` MATCH (start:BenchmarkNode {id: 'bench-100'}) CALL { WITH start MATCH path = (start)-[:RELATED_TO*1..2]-(connected) RETURN connected LIMIT 50 } RETURN DISTINCT connected `); }, { iterations: 50 }); bench('Pattern matching with multiple conditions', async () => { await session.run(` MATCH (n:BenchmarkNode)-[r:RELATED_TO]->(m:BenchmarkNode) WHERE n.value < 500 AND m.value > 500 AND r.weight > 0.5 RETURN n, r, m LIMIT 100 `); }, { iterations: 100 }); bench('Union query', async () => { await session.run(` MATCH (n:BenchmarkNode) WHERE n.value < 100 RETURN n.id, n.value, 'low' AS category UNION MATCH (n:BenchmarkNode) WHERE n.value >= 900 RETURN n.id, n.value, 'high' AS category ORDER BY value LIMIT 50 `); }, { iterations: 100 }); bench('Collect and unwind pattern', async () => { await session.run(` MATCH (n:BenchmarkNode) WHERE n.value % 100 = 0 WITH collect(n) AS nodes UNWIND nodes AS node MATCH (node)-[:RELATED_TO]->(connected) RETURN node.id, collect(connected.id) AS connections LIMIT 25 `); }, { iterations: 50 }); });